Artificial Intelligence: Adoption and Return on Investment

Artificial Intelligence: Adoption and Return on Investment

by Bruno Accioly – 25.09.2025

What MIT Did Not Say

In recent weeks, headlines have trumpeted that "95% of AI projects fail" or that "companies obtain no return whatsoever from the technology." These headlines went viral and were widely repeated by outlets such as Fortune, Forbes, and others. The problem is that they do not correspond to what the MIT report actually shows. At the very least, there was journalistic recklessness; perhaps even incompetence or negligence, since the biased interpretation seems intended to discredit AI as a whole.

The study in question, "The GenAI Divide: State of AI in Business 2025," analyzed 52 companies and more than 300 initiatives. The "95%" figure does not refer to AI projects in general, but to a specific subset: custom corporate tools, expensive and ineffective wrappers that companies attempted to build or purchase. LLMs such as ChatGPT or Gemini, by contrast, showed high rates of adoption and continued use. What spread was not science, but a misread headline.

It is essential to open this article by making this clear: the failure is not AI's, much less that of LLMs. The failure lies in how companies and the press handled the data. On one side, executives buying inflated promises from vendors. On the other, journalists repeating numbers without examining the tables. The result is noise, not information.

What the Study Actually Shows

The MIT report distinguishes between two universes. The first is widespread use: 80% of the companies surveyed have already experimented with general-purpose LLMs — ChatGPT (OpenAI), Claude (Anthropic), Gemini ( DeepMind) — and 40% continue to use them in an integrated manner. In this field, adoption is high and satisfaction is consistent. The second, more restricted universe consists of those specific corporate projects, such as tools sold as panaceas, magical solutions often costing tens of thousands of dollars but delivering less value than a US$20-per-month ChatGPT subscription.

Wes Roth was among those who clearly explained this point: the 5% success rate does not refer to the LLMs that have transformed workers' daily lives, but to highly customized projects, most of which failed. At the same time, there is a parallel economy, what the report calls Shadow AI: employees who, without waiting for IT approval, use personal accounts to automate substantial parts of their work. The contrast is striking: while headlines speak of absolute failure, the workplace is already experiencing diffuse and invisible yet consistent success.

This dismantles the idea that AI does not deliver value. What does not deliver value are poorly directed investments that purchase dubious solutions instead of training people and taking advantage of the LLMs already available. The error lies in the investment and, obviously, in the sensationalist headlines… not in the technology.

The Fraud of Sensationalism

There is, therefore, an "interpretive fraud" underway. It is not that MIT was mistaken; rather, the way media outlets carelessly translated the findings completely distorted the research. Headlines ignored basic distinctions among categories of AI, lumped LLMs together with ineffective wrappers, and concluded that "AI fails." That is absolutely false.

As Matthew Berman emphasized, AI is not "killing the economy"; on the contrary, it is deeply integrated into it. What destroys value are inflated promises, opportunistic consulting firms, and poorly executed implementations. By inflating a decontextualized number, the press does more than make a mistake: it irresponsibly contributes to undermining trust in a technology that is already delivering daily results.

It is no coincidence: negative headlines sell more clicks. It is one of the most absurd conflicts of interest facing the Fourth Estate, which should be responsible for keeping professionals and companies better informed. Journalistic responsibility must not be ignored. By spreading false or poorly contextualized statistics, these outlets fuel misguided skepticism and may stall investments that, if well directed, would bring real gains. This deserves to be denounced as recklessness and negligence.

The War of the Currents

At the end of the nineteenth century, we witnessed a classic episode of technological sensationalism: the so-called War of the Currents. Thomas Edison, already established as an inventor and entrepreneur, defended direct current (DC), while Nikola Tesla and George Westinghouse backed alternating current (AC).

To protect his interests, Edison launched a smear campaign against AC, exploiting public fear. He had public demonstrations organized in which animals were electrocuted with alternating current, attempting to associate Tesla's technology with risk and death. Even the first electric chair was designed in this context as a way to stigmatize AC.

Time showed, however, that alternating current was more efficient, scalable, and suitable for carrying electricity over long distances. It was Tesla's technology that prevailed — despite Edison's campaign of fear. The historical lesson is clear: when economic interests encounter disruptive innovation, sensationalism often overrides technical analysis. Today, apocalyptic headlines about AI repeat, on another stage, the same strategy of fear and discredit that Edison used against Tesla.

Between Automation and Collaboration

One of the study's strongest findings is that the projects that work are not those that try to replace people, but those that expand their capabilities. Full automation may appear efficient, but it tends to generate frustration and limited results. Human-machine collaboration, especially through LLMs, shows high adoption rates and implicit ROI.

"Shadow AI" reinforces this: workers themselves create value by using simple tools to reduce bureaucracy and accelerate workflows. These results are invisible in financial reports but powerful in practice. The paradox is that many companies try to suppress this use in the name of "security" or "control," while pouring millions into poorly conceived corporate projects.

The lesson is clear: AI should not be viewed as a replacement, but as an ally. Those who try to automate everything fail. Those who adopt it collaboratively prosper.

The Scale Divide

Another central point: pilots work, but scaling is difficult. And here again, a superficial reading gets in the way. When it is said that only 5% of projects achieved ROI, it must be remembered that these projects are not open LLM pilots, but custom corporate tools. Even in these cases, some achieved significant back-office gains: multimillion-dollar reductions in outsourcing contracts and savings in marketing and services.

The problem is not the absence of value, but poorly formulated expectations. Companies that enter AI as though purchasing a miracle inevitably become disappointed. Scaling requires strategy, not merely budget. Without organizational and cultural design, any pilot loses its way.

The false narrative of failure therefore conceals the real lessons: where there is clarity of objectives and careful integration, there is return. Where there is the illusion of a miracle, there is abandonment.

The Missing Link: Strategy, Not Tools

The insistence on outsourcing solutions or buying black boxes explains much of the problem. What companies lack is not technology, but decision architects. The role of an AI Strategist, someone capable of defining where to apply AI, how to measure value, and how to integrate workflows, is essential. Without one, every investment becomes a shot in the dark.

The MIT data reinforces this absence: internal projects fail more often than those developed in partnership with experienced vendors. But even partnerships do not solve the problem on their own. What matters is the ability to think about AI strategically, not merely to acquire it as off-the-shelf software.

While the headlines discuss "failure," the reality is different: there is a lack of governance, a lack of clarity, and a lack of prepared people. This cannot be corrected through occasional consulting, but through continuous training.

Technology ROI: 200 Years of Illusions and Delayed Returns

The debate over "return on investment" did not begin with AI. Every transformative technology has gone through a cycle of hype, frustration, and maturation — in which real ROI took time to emerge.

During the Industrial Revolution, the first mechanical looms faced fierce resistance from the Luddites and showed low initial efficiency. ROI came years later, when entire processes were redesigned. The same happened with electricity: during the first 20 years, factories merely replaced steam engines with electric motors, without significant gains. Only when industrial architects reinvented factory layouts to take advantage of flexible wiring and electric lighting did productivity soar.

In the twentieth century, history repeated itself with ERP systems and later with CRM: extremely expensive projects, promises of total integration, and very high failure rates. But when companies learned to redesign processes and train people, ROI appeared and these systems became indispensable.

The pattern is recurrent: the value of a technology lies neither in the pilot nor in the initial hype, but in the profound reorganization it requires. Measuring ROI too early always generates frustration. Measuring it with strategic vision reveals transformative returns. AI is merely going through another chapter in this 200-year historical cycle.

Beyond the Hype Lies Training

The conclusion of this debate is clear. There was no "95% failure rate for AI." There was misreading, bad faith, and sensationalism. AI, especially through LLMs, is already ubiquitous, efficient, and productive. The real failure lies in how companies buy magical solutions and how the press reports on them.

narra positions itself in this landscape neither by selling miracles nor through consulting that imposes outside solutions. We act as a training partner, developing a select group of strategists within companies themselves, an elite task force dedicated to the conception and design of implementation and deployment — or even larger teams, through partners who facilitate Corporate Training. We believe that true ROI comes from prepared professionals, not from US$50,000 wrappers that do less than a mainstream chatbot.

The future of AI will not be defined by hype or sensationalism, but by those who know how to train their teams to use the technology critically, collaboratively, and strategically. This is the necessary turning point. This is our proposal.by Bruno Accioly – 25.09.2025